A novel adaptive-weight ensemble surrogate model base on distance and mixture error

Area of Science:

• Engineering
• Computational Science
• Machine Learning

Background:

• Surrogate models approximate computationally expensive simulations in design optimization.
• Developing accurate surrogate models with limited data is a significant challenge.

Purpose of the Study:

• To propose a novel adaptive-weight ensemble surrogate modeling method.
• To enhance the accuracy of surrogate models, particularly when dealing with limited sample data.

Main Methods:

• An adaptive-weight ensemble approach is developed, considering prediction sample location, a mixture error metric, and component model learning characteristics.
• The method moves beyond single error metrics for improved ensemble weighting.

Main Results:

• The proposed method was tested on five nonlinear benchmark functions and a finite element model of an automotive exhaust pipe.
• Comparative results demonstrated the effectiveness of the ensemble models in achieving higher accuracy.

Conclusions:

• The novel adaptive-weight ensemble surrogate modeling method effectively addresses the challenge of building high-accuracy models with limited samples.
• The approach shows promising potential for improving accuracy in engineering design optimization and simulation analysis.